Abstract
Problem framing
Health systems in Madagascar, Kenya, and Eritrea operate under different governance models, but share a structural constraint: clinical decisions, stock movement, and surveillance signals must often be recorded far from reliable connectivity and specialist backup. National averages hide county, island, and highland variance that breaks cloud-only product assumptions.
This brief synthesizes published workforce, mortality, expenditure, and connectivity indicators to argue for offline-first, ontology-backed health software. The objective is not to replace ministries or NGOs. It is to show where a product studio can ship inspectable systems that reduce data loss, referral delay, and stock-out blind spots in the most underserved catchment areas.
Hypothesis
Software leverage is highest where physician density is lowest and mobile access is uneven.
Trend lines approximate published WHO/UNICEF series for comparative framing. Latest anchor points match GHO country profiles used in the tables below.
Kenya shows higher digital readiness but persistent rural access gaps. Madagascar and Eritrea sit in the quadrant where offline software, CHW tooling, and sync discipline matter most.
Country profiles
Three contexts, one design problem
| Country | Population | Rural share | Physicians / 10k | U5MR | Health spend (% GDP) |
|---|---|---|---|---|---|
| Madagascar | 30.3M | 63% | 0.21 | 51.2 | 3.8 |
| Kenya | 55.2M | 70% | 1.6 | 35.5 | 4.9 |
| Eritrea | 3.6M | 59% | 0.76 | 41.8 | 4.2 |
MDG
Madagascar
Burden: Malaria, maternal mortality, tuberculosis, and malnutrition remain primary outpatient and inpatient drivers.
Connectivity: Low national broadband penetration; mobile-first access in coastal and urban corridors.
Constraint: Island geography, seasonal road disruption, and thin specialist coverage outside Antananarivo.
KEN
Kenya
Burden: HIV/TB co-infection, NCD growth in urban centers, and county-level referral bottlenecks.
Connectivity: Stronger mobile money and API ecosystem; uneven last-mile connectivity in arid north and remote counties.
Constraint: Devolved health system creates data fragmentation across 47 counties without shared ontology.
ERI
Eritrea
Burden: Maternal health, vaccine cold chain, and infectious disease surveillance across dispersed highland communities.
Connectivity: Limited public internet infrastructure; facility reporting often manual and batch-oriented.
Constraint: Sparse specialist network and high cost of real-time sync across remote clinics.
Methods
Research questions
RQ1
Which clinical and logistics signals can be captured offline at community level without breaking auditability?
RQ2
How do referral pathways differ across Madagascar, Kenya, and Eritrea when transport, connectivity, and county boundaries change?
RQ3
What minimum viable ontology lets a national program, NGO partner, and district clinic share the same patient and stock vocabulary?
RQ4
Where can bounded agentic automation reduce clerical load for nurses and CHWs while keeping human review on high-risk decisions?
System design
Product layers that match frontier health operations
Community capture
Offline-first mobile forms for vitals, symptoms, stock counts, and referrals with conflict-safe sync when connectivity returns.
Facility registry
Canonical facility, staff role, and service capability model so dashboards do not double-count mobile clinics or seasonal camps.
Supply and cold chain
Temperature excursion alerts, batch traceability, and reorder logic tuned for irregular delivery schedules.
Surveillance layer
Case aggregation with geospatial clustering for malaria, TB, and outbreak signals without exposing identifiable records in open views.
Review queue
Human-in-the-loop triage for abnormal vitals, stock-outs, and duplicate registrations before escalation to district teams.
Partner export
DHIS2-aligned or CSV export paths for ministries and donors that require standardized reporting windows.
Stratir delivery
Building software for regions the default stack ignores
Stratir applies the same studio discipline used for intelligence and agentic products to public health and humanitarian contexts: ontology first, bounded automation, human review on high-risk paths, and exports partners can audit. The goal is production software that survives sparse connectivity, shared devices, and multi-org reporting windows.
Discuss a health systems build01
Epidemiologic scoping
Stratir maps the decision owners, indicator definitions, facility types, and reporting cadence before any interface is designed.
02
Ontology for sparse networks
Entities (patient, encounter, stock lot, referral, CHW route) are modeled once so county, island, and national views stay aligned.
03
Offline-capable product
Engineering prioritizes low-bandwidth sync, durable local storage, and role-based access on shared devices.
04
Pilot with measurable endpoints
Pilots track time-to-referral, stock-out duration, data completeness, and supervisor review turnaround, not vanity usage metrics.
References
Sources and indicators
WHO Global Health Observatory
Physician density, under-five mortality, and health workforce indicators.
World Bank World Development Indicators
Health expenditure (% of GDP), rural population share, and connectivity proxies.
UNICEF Data
Child mortality and immunization coverage benchmarks for cross-country comparison.
Africa CDC / ministry reporting frameworks
Regional surveillance and export conventions for partner alignment.
Indicator values in tables and charts are anchored to publicly reported WHO GHO and World Bank WDI country profiles for comparative research framing. They should be validated against the latest ministry and partner datasets before operational or procurement use. This brief is a systems research artifact, not clinical guidance.